AWS and Cloudflare Build Tools to Help AI Agents Do Real Work

Amazon Web Services and Cloudflare have both released new features aimed at making AI agents more practical for businesses. Both updates signal a shift from simple, experimental AI projects toward systems that can handle real production work.

AWS expanded its Bedrock AgentCore platform with new browser capabilities, including proxy configuration, browser profiles, and browser extensions that let agents interact with websites the way humans do. The company also launched an AI activity dashboard in AWS WAF—a monitoring tool that helps security teams see what AI bots and agents are doing on company networks.

Meanwhile, Cloudflare positioned itself as a full development platform by launching the industry's first remote Model Context Protocol (MCP) server, making Durable Objects available for free to developers, and releasing a tool called Durable Workflows for production use—all designed with AI agents in mind.

AWS Expands What Agents Can Do

Amazon Bedrock AgentCore is now generally available as a platform for businesses to deploy AI agents in production. The new browser capabilities are the big addition here.

Until now, most AI agents interacted with systems through APIs—essentially, direct, programmed connections. But many enterprise systems (older websites, internal tools, legacy applications) don't have APIs. They were built for humans to click through. The new browser features let an agent navigate these interfaces exactly like a person would: it can use proxy servers, maintain user profiles, and even load browser extensions. This solves a real problem: it lets agents work with systems that haven't been updated to modern integration methods.

The AWS WAF dashboard adds another important piece. Security teams can now see agent traffic separately from human traffic on their networks. This matters because companies increasingly have both humans and AI agents using the same systems. Being able to distinguish them makes it easier to set security rules and spot problems.

Cloudflare Removes Friction for Developers

Cloudflare's approach focuses on making it easier for developers to build agents. The remote MCP server (MCP stands for Model Context Protocol—a standard, developed by Anthropic, for how AI systems talk to outside tools and data) lets developers plug agents into external systems without managing their own servers.

The company also made Durable Objects—a tool for storing information that persists across multiple agent interactions—available for free. Previously this had a cost, which discouraged people from experimenting. Durable Objects are useful because agents often need to remember state across multiple steps. If an agent handles a customer order that takes several back-and-forth exchanges, it needs to remember what happened in each step.

Durable Workflows, now ready for production, handles the orchestration of complex agent tasks. In practice, this means managing what happens when one step finishes and the next begins, what happens when something breaks, and how to retry failed tasks. These are the nuts-and-bolts reliability problems that matter once agents run real business processes.

Cloudflare also updated its Workers AI platform to act as a unified interface for AI models from multiple providers—OpenAI, Anthropic, and others. The benefit is practical: developers can switch which AI model powers an agent without rewriting all the code that talks to it.

Making Development Itself More Efficient

Both platforms integrated with developer tools. Cloudflare updated its prompt engineering helpers for use in editors like Cursor, Windsurf, and Zed, and with ChatGPT and Claude. Now a developer can describe what they want their agent to do in plain language, and these tools generate the initial code automatically.

This reflects a broader trend: cloud providers are optimizing for a world where developers themselves use AI to write AI-powered code. The development environment itself has become agentic.

Why This Matters Now

We have seen this pattern before. When cloud providers first went mainstream, they offered basic compute and storage. But as companies started using them seriously, providers added specialized services tailored to mobile apps, then container orchestration, then serverless functions. Each shift happened when the technology matured from experimental to essential.

The current additions—browser automation, traffic monitoring, state management—suggest that AI agents are moving from proof-of-concept trials toward actual production work. These are the kinds of features that only matter once systems are handling real business processes at scale. Early experiments don't need sophisticated traffic monitoring or complex workflow orchestration. Production systems do.

When infrastructure providers start optimizing for production deployments rather than just making a technology available, it is usually a sign that the broader ecosystem is ready for real adoption. The focus on developer experience and reduced friction also indicates a shift: agent development is moving from research and experimentation toward practical engineering. That is a meaningful moment.

For organizations considering AI agents for their business, these updates provide the technical building blocks for production use. Features like browser controls, network visibility, and state management address the hard operational problems that only emerge once you are running agents at enterprise scale and integrating them with existing systems.